China’s largest technology companies — including Alibaba, Tencent, and Baidu — are accelerating efforts to replace Nvidia chips with homegrown alternatives as US export restrictions bite deeper into their artificial intelligence ambitions, according to a report by the Financial Times.
The shift marks a significant moment for China’s AI sector, as companies scramble to maintain momentum in model training and deployment while navigating increasingly strict US trade controls. A new ban imposed last month by the Trump administration further tightened access to Nvidia’s H20 processors — already a pared-down version built to comply with earlier Biden-era regulations.
Insiders told the Financial Times that current stockpiles of Nvidia chips may only last until early 2026, sparking urgency among Chinese firms to adopt domestic options. However, transitioning is neither simple nor swift: new chip orders take three to six months to arrive, and compatibility challenges loom large, especially when moving from Nvidia’s CUDA ecosystem to local alternatives like Huawei’s CANN software stack.
Baidu, Alibaba, and Tencent have all confirmed they are actively testing domestic chips. “We believe that over time, domestically developed self-sufficient chips… will form a strong foundation for long-term innovation in China’s AI ecosystem,” said Shen Dou, head of Baidu’s AI cloud group. Alibaba CEO Eddie Wu echoed this, noting the company is “exploring diversified solutions,” while Tencent president Martin Lau highlighted efforts to optimize current chip usage and prepare for potential alternatives.
Huawei’s Ascend chips are emerging as a front-runner, with growing adoption across state-owned firms and sensitive sectors. However, fears of US retaliation have made some private companies cautious, particularly after Washington warned that use of restricted chips “anywhere in the world” could trigger penalties.
Other Chinese chipmakers such as Cambricon and Hygon are also being considered, while Alibaba and Baidu continue developing their own processors. Despite the push, tech leaders acknowledge that switching systems could disrupt AI development for up to three months due to the complexity of migrating software frameworks.
According to analysts at GF Securities, Nvidia may introduce a new China-compliant chip in July, but it is expected to lack key features such as high-bandwidth memory, potentially weakening its performance against Chinese rivals.
As US-China tech tensions deepen, the push toward self-reliance is no longer optional. Chinese companies are increasingly adopting a hybrid strategy — using existing Nvidia chips for training while relying on local alternatives for inference — a shift that could redefine the country’s AI landscape in the years ahead.
